移動通信大數據分析——數據挖掘與機器學習實戰 Mining Over Air: Wireless Communication Networks Analytics

本書以4G/5G無線技術、機器學習和數據挖掘的新研究和新應用為基礎，對分析方法和案例進行研究；從工程和社會科學的角度，提高讀者對行業的洞察力，提升運營商的運營效益。本書利用機器學習和數據挖掘技術，研究移動網絡中傳統方法無法解決的問題，包括將數據科學與移動網絡技術進行完美結合的方法、解決方案和算法。 本書可以作為研究生、本科生、科研人員、移動網絡工程師、業務分析師、算法分析師、軟件開發工程師等的參考書，具有很強的實踐指導意義，是不可多得的專業著作。

第一作者簡介
歐陽曄博士
亞信科技首席技術官、高級副總裁
歐陽曄博士目前全面負責亞信科技的技術與產品的研究、開發與創新工作。
加入亞信科技之前，歐陽曄博士曾任職於美國第一大移動通信運營商威瑞森電信（Verizon）集團，擔任通信人工智能係統部經理，是威瑞森電信的Fellow。
歐陽曄博士在移動通信領域擁有豐富的研發與大型團隊管理經驗，工作中承擔過科學家、研究員、研發經理、大型研發團隊負責人等多個角色。
歐陽曄博士專注於移動通信、數據科學與人工智能領域跨學科研究，致力於5G網絡智能化、BSS/OSS融合、通信人工智能、網絡切片、MEC、網絡體驗感知、網絡智能優化、5G行業賦能、云網融合等領域的研發創新與商業化。

第1章概述
1.1 電信業大數據分析···························1
1.2 電信大數據分析的驅動力················2
1.3 大數據分析對電信產業價值鏈的益處··················································3
1.4 電信大數據的實現範圍····················4
1.4.1 網絡分析···················································5
1.4.2 用戶與市場分析·······································8
1.4.3 創新的商業模式·······································9
1.5 本書概要··········································9
參考文獻·················································10

第2章電信分析方法論
2.1 回歸方法········································12
2.1.1 線性回歸··················································13
2.1.2 非線性回歸··············································15
2.1.3 特徵選擇··················································16
2.2 分類方法········································18
2.2.1 邏輯回歸··················································18
2.2.2 其他分類方法··········································19
2.3 聚類方法········································20
2.3.1 K均值聚類··············································21
2.3.2 高斯混合模型··········································23
2.3.3 其他聚類方法··········································24
2.3.4 聚類方法在電信數據中的應用·················25
2.4 預測方法········································25
2.4.1 時間序列分解··········································26
2.4.2 指數平滑模型··········································27
2.4.3 ARIMA模型············································28
2.5 神經網絡和深度學習·····················29
2.5.1 神經網絡··················································29
2.5.2 深度學習··················································31
2.6 強化學習········································32
2.6.1 模型和策略··············································33
2.6.2 強化學習算法··········································33
參考文獻·················································34

第3章LTE網絡性能趨勢分析
3.1 網絡性能預測策略·························39
3.1.1 直接預測策略··········································39
3.1.2 分析模型··················································39
3.2 網絡資源與性能指標之間的關係···40
3.2.1 LTE網絡KPI與資源之間的關係···········40
3.2.2 回歸模型··················································41
3.3 網絡資源預測·································43
3.3.1 LTE網絡流量與資源預測模型···············43
3.3.2 預測網絡資源··········································43
3.4 評估RRC連接建立的應用············46
3.4.1 數據準備與特徵選取······························46
3.4.2 LTE KPI與網絡資源之間的關係推導····47
3.4.3 預測RRC連接建立成功率·····················49
參考文獻·················································50

第4章熱門設備就緒和返修率分析
4.1 設備返修率與設備就緒的預測策略················································53
4.2 設備返修率和就緒預測模型··········54
4.2.1 預測模型的移動通信服務························54
4.2.2 參數獲取與存儲······································55
4.2.3 分析引擎··················································56
4.3 實現和結果·····································58
4.3.1 設備返修率預測······································58
4.3.2 設備就緒預測··········································62

第5章VoLTE語音質量評估
5.1 應用POLQA評估語音質量··········68
5.1.1 POLQA標準···········································68
5.1.2 語音質量評價中的可擴展性和可診斷性··················································69
5.2 CrowdMi方法論····························69
5.2.1 基於RF特徵的分類·······························70
5.2.2 網絡指標選擇與聚類······························70
5.2.3 網絡指標與POLQA評分之間的關係····70
5.2.4 模型測試··················································70
5.3 CrowdMi中的技術細節·················71
5.3.1 記錄分類··················································71
5.3.2 網絡指標的選擇······································71
5.3.3 聚類·························································72
5.3.4 回歸·························································73
5.4 CrowdMi原型設計與試驗·············74
5.4.1 客戶端和服務器架構······························74
5.4.2 測試和結果··············································76
參考文獻·················································78

第6章移動APP無線資源使用分析
6.1 起因和系統概述·····························80
6.1.1 背景和挑戰··············································80
6.1.2 移動資源管理··········································81
6.1.3 系統概述··················································82
6.2 AppWiR眾包工具··························83
6.3 AppWiR挖掘算法··························84
6.3.1 網絡指標的選擇······································84
6.3.2 LOESS方法············································87
6.3.3 基於時間序列的網絡資源使用預測·······87
6.4 實現和試驗·····································88
6.4.1 數據收集與研究······································88
6.4.2 結果和準確度··········································89
參考文獻·················································91

第7章電信數據的異常檢測
7.1 模型················································93
7.1.1 高斯模型··················································94
7.1.2 時間依賴的高斯模型······························94
7.1.3 高斯混合模型（GMM）·························95
7.1.4 時間依賴的高斯混合模型·······················95
7.1.5 高斯概率潛在語義模型（GPLSA）·······95
7.2 模型對比········································97
7.2.1 樣本定義··················································97
7.2.2 異常識別··················································98
7.2.3 時間依賴GMM與GPLSA的對比·········99
7.3 仿真與討論···································100
參考文獻···············································103

第8章基於大數據分析的LTE網絡自優化
8.1 SON（自組織網絡）···················105
8.2 APP-SON ······································107
8.3 APP-SON架構·····························108
8.4 APP-SON算法·····························110
8.4.1 匈牙利算法輔助聚類（HAAC）··········111
8.4.2 單位回歸輔助聚類數的確定·················114
8.4.3 基於DNN的回歸·································114
8.4.4 每個小區在時序空間的標籤組合·········116
8.4.5 基於相似性的參數調整·························116
8.5 仿真與討論···································117
參考文獻···············································122

第9章電信數據和市場營銷
9.2.1 數據採集和數據類型····························130
9.1 電信營銷專題·······························127
9.2.2 網絡的提取和管理································131
9.2 社交網絡的總體構建···················130
9.3 網絡結構的度量···························133
9.4 網絡中的消費者行為建模············134
參考文獻···············································135

第10章傳染式客戶流失
10.1 問題引入·····································138
10.1.1 流失率問題··········································138
10.1.2 社交學習和網絡效應··························139
10.2 網絡數據的處理·························141
10.3 動態模型·····································143
10.3.1 模型介紹··············································143
10.3.2 模型的定義··········································144
10.3.3 自身經驗建模、社交學習和社交網絡效應······································146
10.3.4 模型估計··············································148
10.4 結果············································149
參考文獻···············································151

第11章基於社交網絡的精準營銷
11.1 網絡效應的渠道·························158
11.2 社交網絡數據處理·····················159
11.3 建模策略問題·····························160
11.3.1 線性空間自回歸模式···························160
11.3.2 社交網絡交互模型······························162
11.3.3 內生同伴效應······································162
11.4 發現與應用·································164
11.4.1 結果的解釋··········································164
11.4.2 基於社交網絡的精準營銷···················165
參考文獻···············································168

第12章社交影響和動態社交網絡結構
12.1 動態模型·····································177
12.1.1 連續時間馬爾可夫模型假設···············177
12.1.2 模型估計與識別··································179
12.1.3 網絡結構對社交影響的多元分析·······180
12.2 研究發現總結·····························181
12.2.1 隨機行動者動態網絡模型的估計結果··············································182
12.2.2 元回歸分析結果··································184
12.2.3 策略模擬··············································188
12.3 結論············································193
參考文獻···············································194